True, it wouldn’t be ethical to conduct an experiment, but we can (and probably do) collect lots of observational data that can provide meaningful insight. People are arrested at all stages of CSAM related offenses from just possession, distribution, solicitation, and active abuse.
While observation and correlations are inherently weaker than experimental data, they can at least provide some insight. For example, “what percentage of those only in possession of artificially generated CSAM for at least one year go on to solicit minors” vs. “real” CSAM.
If it seems that artificial CSAM is associated with a lower rate of solicitation, or if it ends up decreasing overall demand for “real” CSAM, then keeping it legal might provide a real net benefit to society and its most vulnerable even if it’s pretty icky.
That said, I have a nagging suspicion that the thing many abusers like most about CSAM is that it’s a real person and that the artificial stuff won’t do it for them at all. There’s also the risk that artificial CSAM reduces the taboo of CSAM and can be an on-ramp to more harmful materials for those with pedophilic tendencies that they otherwise are able to suppress. But it’s still way too early to know either way.